Aravind Srinivas is currently in a competitive struggle with Google to secure a place for his Perplexity AI assistant as a pre-installed application on Android devices. Concurrently, he is redirecting his startup’s focus toward what he believes will be an essential arena in the AI competition: web browsers.
Srinivas announced plans for the launch of Comet, Perplexity’s own web browser, slated for next month. He emphasized, “The reason we’re doing the browser is that it might be the best way to build agents. A browser functions like a containerized operating system, enabling access to various third-party services through hidden tabs, provided you’re logged in. It allows for client-side page scraping and reasoning, ultimately taking actions on your behalf.”
Srinivas is not alone in his pursuits; several AI companies are moving toward integrating browser functionalities into their products. OpenAI’s Operator and Google’s Mariner are examples that utilize the browser to manage commands and navigate websites. Although OpenAI has not introduced a browser yet, there are speculations about its development. On the other hand, the U.S. government may compel Google to divest Chrome due to antitrust rulings concerning its dominance in the search market.
According to one of Srinivas’s associates, Perplexity would be interested in adopting Chrome should it be separated from Google, and OpenAI has also expressed interest in this direction.
The future of Chrome is uncertain, yet the ongoing antitrust investigations of Google have already provided Perplexity with new opportunities for distribution partnerships with manufacturers of Android devices. Recently, Motorola confirmed that Perplexity will come pre-installed on its latest Razr models, potentially increasing access to millions of users for Srinivas’s AI tool. Although he acknowledges the integration is not as extensive as desired by either party, he views this milestone as a significant achievement for the smaller company.
“Had Google not faced legal scrutiny from the DOJ, this partnership would not have materialized,” he remarked. “Google would have intimidated a lot of OEMs. In conversations with telecommunication companies, I encountered reluctance to engage with us due to fears that Mountain View’s displeasure could jeopardize their revenue share.”
During my last discussion with Srinivas over a year ago, Perplexity had approximately 1 million users and had secured under $100 million in funding. Today, the startup boasts nearly 30 million active users monthly and has successfully attracted hundreds of millions of dollars in investment. Srinivas reported that Perplexity is currently processing around 600 million queries each month, which accounts for roughly 14% of Google’s search volume.
The following dialogue with Srinivas occurred just before his announcement with Motorola, touching on his efforts to expand Perplexity’s reach, the strategic importance of owning the browser interface, and additional insights into the development of an iOS assistant designed to control various applications.
This conversation has been edited for length and clarity.
Can you explain how the Motorola partnership was established and the obstacles you faced with Google?
Our talks accelerated upon presenting a demo of the Perplexity Android assistant, which was launched in January. Motorola tested the application and found it to perform significantly better than its rival Gemini, which generated enthusiasm for preloading the app and notifying users to set Perplexity as their default assistant. However, Google intervened, stating that Motorola could not proceed with the phone’s launch using the official version of Android and the Play Store unless Gemini was established as the default.
Without the DOJ trial, our collaboration wouldn’t have been feasible. Google would have pressured many OEMs, and I faced multiple encounters with telecom companies that were unwilling to meet with us due to potential repercussions from Google.”
Changing the default assistant requires several steps, as Google maintains substantial influence over the Android ecosystem.
With Samsung’s investment in Perplexity, do you foresee a partnership with them akin to your collaboration with Motorola?
Absolutely. I’m optimistic about establishing some form of cooperation with Samsung. The specifics surrounding default settings or onboarding processes are still under discussion.
Your focus appears to be heavily on distribution and partnerships to scale Perplexity.
We are eager to collaborate with anyone. We are currently working with telecom companies and are aiming to partner with OEMs as we advance toward browser development, which will also have versions for Mac and Windows. We will seek to establish partnerships in these areas as well.
Similar to Android where Google controls OEM relationships, Microsoft has stringent agreements with laptop manufacturers. It will be a continuous challenge for us in that arena.
You just introduced your assistant on iOS, and user reactions seem very positive regarding its capabilities. Did you receive any special permissions from Apple for app control?
No special permissions were granted. Our system cannot be used for actions such as setting alarms or enabling low power mode. We cannot make phone calls or send iMessages either.
We utilized the Apple EventKit SDK to access Reminders, Podcasts, Apple Music, Apple Maps, and some additional Apple applications, thereby leveraging our search infrastructure alongside deep linking to apps like YouTube and Uber.
While Siri excels at basic tasks such as setting alarms, it struggles to find specific songs or podcasts. We believe we have effectively addressed those needs.
What motivated the development of your own browser, and when can users expect its release?
We decided to create a browser as it might be the most effective means to generate agents. On mobile platforms, we lack OS-level access, hindering our ability to call applications and retrieve their data. While we can deep link, we cannot assess ride prices or compare services like Uber and Lyft directly.
Thus, the development of a browser was necessary as it acts as a functional operating system in itself. It enables easy access to third-party services through hidden tabs and allows for actions and reasoning on behalf of users. This framework appeals to us, as answering questions will soon be commonplace.
You’ve faced criticism from publishers regarding content scraping, though you’ve begun compensating some. Are you satisfied with the current publisher relationships, or is more work needed?
While there is certainly room for improvement, the situation has notably progressed since our last discussion. We adhere to scraping regulations as outlined in robots.txt and only employ third-party data providers for content we cannot scrape.
Reports indicate you are in the process of raising substantial funds at an $18 billion valuation. How do you intend to utilize these funds?
To develop reliable agents, we plan to leverage advanced reasoning models. While today’s costs may seem high, they will decline in the future. However, we cannot postpone our rollout; we need to engage as many users as possible to gather data, refine models, and ultimately reduce expenses.
What’s the current status of your interest in acquiring TikTok? Have you had any recent discussions with the White House regarding funding concerns?
I have not abandoned that goal; my prospects may not have been the strongest. However, there were sufficient investors willing to support me. Concerns from ByteDance were not related to funding but rather their desire to maintain control over the algorithm, believing they manage it better than anyone else.
Separating the app running in the U.S. and Europe is complex due to intertwined operations. Regulatory tariffs will influence everything, including TikTok.
Do you have concerns that the prominence of ChatGPT might deter users from trying Perplexity? ChatGPT’s features enhance user retention through personalization.
It appears their approach, as discussed by Sam Altman in an interview with Ben Thompson, aims to embed a “Login with ChatGPT” feature in third-party applications to consolidate user data. This could be contingent on successfully persuading those developers.
Our strategy emphasizes allowing users to remain logged in where they already are. By developing a browser, we will facilitate easier app access on the user’s behalf.
The key to memory will hinge on the company that possesses the most contextual information. While ChatGPT may have limitations regarding user transaction data, aiding personalization is where we see an opportunity.
Translating user activity and preferences from multiple applications into a unified assistant experience, across platforms, poses a significant challenge. This highlights the necessity of building our browser on both web and mobile devices, enabling seamless sharing among apps.
The browser seems to be a pivotal component of your overarching vision.
There’s potential for future expansion, including developing operating systems for Windows, Mac, Android, or iOS. A browser presents only a limited, containerized environment. The true objective lies in creating a comprehensive OS.
Elsewhere
- OpenAI’s computing requirements: The company’s closed-door investor day in San Francisco underscored the need for greater computing resources to sustain ChatGPT’s rapid expansion, surpassing even profitability concerns. This trend suggests rising costs per token for frontier models despite efficiency improvements. (The Information recently provided a detailed analysis of OpenAI’s financial projections.)
- Quick updates: Elon Musk shared plans to step back from DOGE in May. / Apple and Meta faced fines under the EU’s Digital Markets Act, criticized by the U.S. as “economic extortion.” / Meta has laid off employees in its Reality Labs division. / Intel is restructuring its teams and introducing a four-day work week. / Google is mandating some remote workers to return to the office three days a week.
Personnel log
Notable career movements and job opportunities:
- Discord co-founder and CEO Jason Citron is stepping down to pursue personal passions, including gaming. Humam Sakhnini, a former executive at Activision Blizzard, will replace him as Discord’s CEO as the company plans for an IPO. The future direction concerning Discord’s focus—whether to maintain a gaming-centric approach—remains to be seen.
- Sam Altman stepped down as the chairman of nuclear energy company Oklo to avoid potential conflicts of interest while pursuing AI partnerships. This change comes amidst Oklo’s public trading status which demands more stringent governance.
- Ranjit Desai and a number of other leaders at Apple have been reassigned from the Vision Products Group to concentrate on Siri enhancements.
- DeepSeek is seeking product leaders in China to develop next-generation intelligent experiences.
Link list
Further reading:
- The Gemini app has achieved 350 million monthly users.
- Microsoft Copilot’s user growth has flattened at 20 million weekly users.
- Dario Amodei discusses the “urgency of interpretability” in the AI sector.
- Kevin Systrom shares insights following his return from retirement.
- Explores the struggle for talent retention at startups nearing IPOs like Figma.
- X’s advertising growth is underwhelming, but the platform finds ways to monetize data.
- An inside look at Amazon’s “Project Greenland” effort to bolster GPU availability.
- Phoebe Gates, daughter of Bill and Melinda Gates, has launched an AI shopping app.
- 60 Minutes interviews Google DeepMind CEO Demis Hassabis.
- Exploring the mystery behind Infinite Reality, a metaverse startup valued at $15 billion.
For more updates, consider subscribing to Technology News, which includes unlimited access to Command Line and all our reports.
I invite your feedback, particularly regarding this topic or additional story ideas. Feel free to respond here or reach out securely via Signal.
Thank you for being a subscriber.
